LLMPIDTuner: Paper Evaluation Results and Balanced SFT Dataset

This record provides the reproducibility data associated with the article "A Physics-Informed Framework for PID Tuning of Chemical Processes Using Large Language Model Agents." It contains the final frozen-protocol FOPDT and SOPDT evaluation runs, prompt-ablation and selected control-style results, IMC references, base/SFT/GRPO small-model evaluations, SFT and GRPO training logs and manifests, frozen protocol assets, derived paper figures and tables, and the balanced-only 40,000-row SFT dataset (20,000 FOPDT and 20,000 SOPDT samples). The frozen protocol identifier is `perturbed_imc_delay_stratified_v1`. File-level and archive-level SHA-256 manifests are included. Model weights, optimizer states, API credentials, private infrastructure paths, obsolete mixed-style experiments, and historical collaborator archives are not included. The released implementation and results improve and extend the work developed from Zhoupeng Shou's master's thesis and are not a complete reproduction of the original thesis implementation.